Concurrent task demands and individual differences in the architecture of reading: Discriminating artifacts from real McCoys.

Abstract

1991) observed a striking effect in which low-frequency exception words were pronounced faster under a larger memory load although other types of words slowed down. This effect strongly favored dual-route models of pronunciation. S. E. Bernstein and T. H. Carr (1996) reported that only certain selected readers produced Paap and Noel's effect, suggesting individual differences in reading-system architecture. P. M. Pexman and S. J. Lupker (1998) criticized Bernstein and Carr's rinding as artifactual and failed to replicate it with a different way of selecting readers. They also failed to support an alternative individual-difference hypothesis of their own. In the current article, the authors developed a new approach to identifying individual differences and applied the approach to a variety of data. They found little evidence of systematic individual differences in the occurrence of Paap and Noel's effect, and indeed, they found little evidence that the effect could be replicated at all. Therefore, they gave up.

Paap and Noel revisited: Individual differences in independent word recognition tasks in relation to the dual-task paradigm

“…This pattern is best accommodated within a dual-route framework. As participants were allocated to skill groups on the basis of performance on tasks that were independent of the experimental naming task, then it is unlikely that the observed effects are due to any statistical artifacts (see Bernstein et al, 1998;Paap & Herdman, 1998;Pexman & Lupker, 1998).…”

“…It is suggested that this occurs because for low-frequency irregular words, incongruent information arrives at the phonological buffer simultaneously from the lexical and sub-lexical routes, thus requiring any decision/synthesis mechanism to resolve conflict prior to output. However, there is some evidence that regularity effects can also be obtained with high-frequency words (cf., Bernstein, DeShon, & Carr, 1998;Content, 1991;Jared, 1997;Lupker, Brown, & Colombo, 1997). As single-route models were also able to account for the frequency/regularity interaction, further evidence was needed to support dual-route theories.…”

“…Only the participants exhibiting these criteria showed release from competition and a three-way interaction, although this failed to reach significance by items. Critics of the Bernstein and Carr study argue that they used criteria for allocating participants to groups that were likely to produce release from competition as a result of regression to the mean, rather than any real differences in word recognition architecture (cf., Bernstein et al, 1998;Paap & Herdman, 1998;Pexman & Lupker, 1998). This concept has been rigorously investigated by Pexman and Lupker (1998).…”

“…Paap and Herdman also suggest that the Pexman and Lupker participant pools did not contain a sufficient number of participants for whom word recognition involved a balanced dual-route system (thought to be indicated by a Frequency × Regularity interaction under low load conditions). They suggest that if pre-selection of groups is made from a homogeneous and highly skilled pool, then attempts to select those with a dual-route architecture can only be based on noise and will be more likely to produce spurious release from competition effects because of regression to the mean (see Bernstein et al, 1998;and Paap & Herdman, 1998, for an extended discussion on this issue). Pexman and Lupker and Paap and Herdman suggest that if participants are pre-selected on reading skill measures outside of the naming task used in the dual-task paradigm, then any observed effects are less likely to be due to statistical artifacts and more likely to be due to individual differences in dual-task paradigm performance in relation to the independent reading task performance.…”

Highly skilled participants and failures to redirect attention: Two plausible reasons for failing to replicate Paap and Noel's effect.

“…As Bernstein, DeShon, and Carr (1998) now allow, an analysis that uses M L as a covariate does not protect against all forms of statistical artifact. A representative experiment from Simulation 2 can be used to illustrate the problem.…”